feat: add seq cursor, work_items, provider fallback, and delivery reliability

Major reliability improvements to the message processing pipeline:

- Add messages.seq monotonic cursor replacing timestamp-based cursors,
  preventing message loss from timestamp collisions with LIMIT queries
- Add work_items table separating agent production from delivery,
  enabling delivery retry without re-running the agent
- Propagate Discord send failures instead of silently swallowing them
- Add Claude 429 → Kimi K2.5 automatic provider fallback with cooldown
- Fix follow-up turn state reset in live index.ts path (not just
  message-runtime.ts) to prevent final output from being lost
- Add restart context tracking for graceful restart announcements
- Lazy migration from timestamp cursors to seq cursors for existing data

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/channels/discord.test.ts

@@ -761,7 +761,7 @@ describe('DiscordChannel', () => {
       expect(currentClient().channels.fetch).toHaveBeenCalledWith('9876543210');
     });
 
-    it('handles send failure gracefully', async () => {
+    it('propagates send failure to the caller', async () => {
       const opts = createTestOpts();
       const channel = new DiscordChannel('test-token', opts);
       await channel.connect();
@@ -770,10 +770,9 @@ describe('DiscordChannel', () => {
         new Error('Channel not found'),
       );
 
-      // Should not throw
       await expect(
         channel.sendMessage('dc:1234567890123456', 'Will fail'),
-      ).resolves.toBeUndefined();
+      ).rejects.toThrow('Channel not found');
     });
 
     it('does nothing when client is not initialized', async () => {